National Poverty Center Working Paper Series
#13 - 06
May 2013
Rising Extreme Poverty in the United States and the Response of
Federal Means-Tested Transfer Programs
H. Luke Shaefer, University of Michigan and Kathryn Edin, Harvard
University
This paper is available online at the National Poverty Center Working Paper Series index at:
http://www.npc.umich.edu/publications/working_papers/
Any opinions, findings, conclusions, or recommendations expressed in this material are those of the author(s) and do
not necessarily reflect the view of the National Poverty Center or any sponsoring agency.
Rising Extreme Poverty in the United States and the Response of Federal Means-Tested
Transfer Programs
In press, Social Service Review
H. Luke Shaefer, PhD
Assistant Professor,
University of Michigan School of Social Work
Research Affiliate
National Poverty Center
Kathryn Edin
Professor,
John F. Kennedy School of Government
Harvard University
Senior Research Affiliate
National Poverty Center
Acknowledgements
H. Luke Shaefer’s time spent on this project was funded in part by a cooperative research
contract (58-5000-0-0083) between the National Poverty Center (NPC) at the University of
Michigan and the U.S. Department of Agriculture, Economic Research Service (ERS) Food and
Nutrition Assistance Research Program (FANRP). The ERS project representative is Alisha
Coleman-Jensen. The views expressed are those of the authors and not necessarily those of the
NPC, ERS, or USDA.
The authors thank Sheldon Danziger, Christopher Jencks, Julia Henly, Tim Smeeding, Bridget
Lavelle, and Matthew Rutledge for helpful comments and suggestions on earlier versions of this
work. All mistakes belong to the authors alone.
1
Abstract
This study documents an increase in the prevalence of extreme poverty among U.S. households
with children between 1996 and 2011, and assesses the response of major federal means-tested
transfer programs. “Extreme poverty” is defined using a World Bank metric of global poverty:
$2 or less, per person, per day. Using the 1996-2008 panels of the Survey of Income and
Program Participation (SIPP), we estimate that in mid-2011, 1.65 million households with 3.55
million children were living in extreme poverty in a given month, based on cash income. This
constituted 4.3 percent of all non-elderly households with children. The prevalence of extreme
poverty has risen sharply since 1996, particularly among those most impacted by the 1996
welfare reform. Adding SNAP benefits to household income reduces the number of extremely
poor households with children by 48.0 percent in mid-2011. Adding SNAP, refundable tax
credits, and housing subsidies reduces the number 62.8 percent.
2
Introduction
The 1996 welfare reform ended the only cash entitlement program for poor families with
children, replacing it with a program that offers time-limited cash assistance and requires ablebodied recipients to participate in work-related activities. This reform, combined with changes to
other public programs that have raised the benefits of work for low-income households, has been
followed by a dramatic decline in cash assistance caseloads and a dramatic increase in workconditioned benefits for the working poor (Ben-Shalom, Moffit & Scholz, in press). That is, total
public means-tested transfers have fallen for the non-working poor, but increased for the working
poor.
In the midst of the slow recovery following the Great Recession, millions of parents are
experiencing long spells of unemployment, but they have little immediate access to means-tested
cash income support. Have these changes coincided with the growth of a new group of American
poor: households with children living on virtually no income? We refer to this group as the
“extreme poor”, and adopt one of the World Bank’s measures of global poverty to define it: $2
or less, per person, per day.
This article presents estimates of the descriptive prevalence of extreme poverty among
households with children in the United States between 1996 and 2011. We expect that extreme
poverty increased between 1996 and 2011, at least in part because of the 1996 welfare reform
and the Great Recession. While this investigation is motivated by these two key factors,
estimates presented here do not fully assess the casual mechanisms of any such rise in extreme
poverty. However, this descriptive documentation is an important first step to understanding
extreme poverty in the U.S., which to our knowledge has never been measured in this way. Since
1996, cash assistance receipt has declined precipitously. In contrast, the percentage of low-
3
income households that receive means-tested benefits from the Supplemental Nutrition
Assistance Program (SNAP) and means-tested refundable tax credits (particularly the Earned
Income Tax Credit) has increased. We hypothesize that these means-tested benefits expansions,
particularly of SNAP, have partially offset the increase extreme poverty since 1996.
Background
Several factors may be related to recent changes in the incidence of extreme poverty in
the U.S. First is the 1996 welfare reform, which replaced a need-based entitlement program, Aid
to Families with Dependent Children (AFDC), with a more restrictive federal block grant
program called Temporary Assistance for Needy Families (TANF). Subsequently, cash
assistance caseloads have fallen from 12.3 million recipients per month in 1996 to 4.5 million in
December 2011, and only 1.1 million of these beneficiaries are adults. Even during the current
period of continued high unemployment, the cash assistance rolls have increased only slightly.
‘Welfare,’ in the form of cash assistance, is a shell of its former self.
Other means-tested income support programs have grown considerably, most notably the
Supplemental Nutrition Assistance Program (SNAP, formerly the Food Stamp Program), and the
Earned Income Tax Credit (EITC), along with other refundable tax credits for low-income
families.1 Relaxed eligibility requirements combined with increased need during the Great
Recession has led SNAP caseloads to increase from an average of 25.5 million recipients per
month in 1996 to 47.5 million in October 2012. The number of recipient families claiming the
EITC rose from 19.5 million in 1996 to 27.8 million in tax year 2010 (Tax Policy Center, 2010).
While the EITC is restricted to wage earners, there is some evidence that many recipients save
part of it throughout the year to protect themselves during short periods with no income
1
Another example of in-kind benefits expansions is increased accessibility of public health insurance for lowincome children and, in some states, their parents.
4
(Mendenhall, Edin, Crowley, Sykes, Tach, Kriz & Kling, 2012). Housing subsidies also are
accessed by a moderate percentage of low-income families.
The 2009 American Recovery and Reinvestment Act (ARRA) temporarily increased
SNAP benefits by about 15 percent per household. At the same time, the EITC was expanded
considerably for households with three or more children and the child tax credit was expanded
and extended to families with lower earnings levels. These major expansions in means-tested aid
were responses to the greatest economic contraction in modern times, which may have had its
own effect on extreme poverty in the U.S. While the Great Recession era has ushered in a
prolonged period of high unemployment, its real legacy may prove to be the unprecedented
duration of unemployment spells: the average spell was 38.1 weeks as of December 2012.
This period has also seen an unprecedented expansion of Unemployment Insurance (UI)
eligibility durations, up to 99 weeks for some workers. However, recent studies find that lowearning workers continue to experience significant difficulty accessing UI benefits, largely as a
result of ineligibility related to the reason for employment separation, as well as possibly
confusion about eligibility (Gould-Werth & Shaefer, 2012; O’Leary & Kline, 2008; Shaefer,
2010). Thus, a rise in the number of households experiencing prolonged periods of
unemployment may have also led to a rise in the number of households surviving on virtually
nothing.
In addition to reporting the official poverty rate, the Census Bureau also reports the “deep
poverty” rate. Those in deep poverty have incomes below 50 percent of the official poverty
threshold. Over the past two decades, deeply poor families have seen a substantial decline in the
amount of aid they receive from public programs, while low-income families between 50 and
150 percent of poverty have actually realized an increase in assistance from public programs, on
5
average (Ben-Shalom, Moffitt & Scholz, in press). But even deeply poor families may have been
affected by changes to public programs and the economy in the last 15 years in heterogeneous
ways. A single mother with a regular minimum wage job working an average of 25 hours a week
would still be in deep poverty but could receive significant EITC and child tax credit benefits,
may access child care subsidies, and could remain on SNAP. In contrast, a family in which the
head has multiple barriers to work and is unable to maintain regular employment will receive no
or at most a small EITC and child tax credit, likely will not be able to access unemployment
insurance benefits because of inadequate labor force attachment, could not access child care
subsidies, and is much less likely to access cash assistance than before the 1996 reform. Thus,
alternative metrics may be necessary to capture the changing circumstances of the poorest of the
poor in the U.S.
Several studies have documented that an increasing proportion of single-mothers are
“disconnected,” with neither earnings nor welfare (Blank, 2007; Loprest, 2011; Turner, Danziger
& Seefeldt, 2006). As much as one-quarter of all single mothers were disconnected for at least a
four-month period in 2009 (Loprest, 2011). Turner et al. (2006) find that these disconnected
mothers often face multiple barriers to work such as learning disabilities, physical limitations,
few work skills, and mental health problems. While these studies offer critical information about
single mother families, other poor families with children also may have been impacted by the
retrenchment of cash assistance and the current period of sustained high unemployment.
To better capture the full extent of a household’s resources, it is important to account not
just for a family’s household earnings and cash welfare, but also other sources of unearned and
in-kind income, such as loans or gifts from family and friends, SNAP, public housing, and EITC
benefits accrued during previous spells of work, even among the jobless.
6
The current article offers an alternative metric designed to measure the prevalence of
extreme destitution among households with children between 1996 and 2011, attempting to
account for most sources of income as comprehensively as possible, including cash and in-kind
income from public programs, informal work, and gifts and loans from family and friends inside
and outside the household. To our knowledge, no previous study has examined U.S. poverty at
such an extreme level.
The primary goal of this analysis is to identify whether there has been an increase in the
number and proportion of households with children living on virtually no income that has
coincided with the implementation of the 1996 welfare reform, particularly in the context of
Great Recession period. This descriptive analysis is motivated by the major retrenchment of cash
assistance and the current period of prolonged economic hardship, but the estimates presented
here are not sufficient to make causal claims about the impact of these factors in particular. More
advanced analyses would be required to make such causal claims.
The measure of “extreme poverty” used here is based on one of the World Bank’s key
indicators of global poverty: $2 per person, per day.2 Tellingly, the World Bank does not release
official estimates for the United States for this metric because it is meant to capture poverty
based on “the standards of the poorest countries” (Ravallion, Chen & Sangraula, 2008, p. 3).
Living below this metric is widely considered to be a marker of extreme destitution, which is
assumed to be very uncommon among wealthy nations. Thus, this metric is conceptually
appropriate for the current paper because the goal is to examine the incidence of extreme
destitution in the U.S., and how it has changed over the study period. Because the current
analysis examines a sub-population (of households with children), and because our estimates and
methodology are only modeled after but are not identical to what is used by the World Bank,
2
http://data.worldbank.org/indicator/SI.POV.GAP2.
7
direct comparison of the estimates presented here with the World Bank’s estimates from
developing countries is not the focus of the current article.
Households are counted as being in extreme poverty if they report $2 dollars or less per
person, per day in total household income in a given month (approximated as $60 per person, per
month in 2011 dollars). In sensitivity analyses, we examine our estimates using a few other
income thresholds. These are reported in table 3 and discussed later on. The official poverty line
for a family of three would equate to roughly $17.00 per person, per day, averaged over a year,
so our measure is roughly 13 percent of the official poverty line. The threshold for deep poverty
(half the poverty line) would equate to an average of approximately $8.50, per person, per day,
over four times the cut-off we use for extreme poverty. We hypothesize that the prevalence of
extreme poverty has risen since 1996, but that this increase has been mitigated by the expansions
of the means-tested aid, particularly of SNAP and the EITC.
Data and Method
We analyze data from the Survey of Income and Program Participation (SIPP), collected
by the U.S. Census Bureau. Our study period starts in 1996, before states were required to
implement the 1996 welfare reform, and before the national unemployment rate fell to a low of
4.0 percent in 2000. The period ends with the most recent available SIPP data, from the middle
of 2011, when the national unemployment rate was roughly 9 percent.
We use a comprehensive household-level total monthly income measure that includes the
resources of all individuals living in a housing unit. This income variable includes labor market
earnings, pension and retirement benefits, cash income from public programs such as
AFDC/TANF (but not in-kind transfers), asset income (dividends, interest and rents) and
reported income from friends and family members outside the household (including child
8
support), and from informal sources (such as odd jobs). We restrict our sample to households
with children under 18 and with household heads under 65. We also drop observations in which
households report negative income values (which may be related to investments). These tend to
be from households with higher incomes in other months. We adjust income values to 2011
dollars using the Consumer Price Index for urban customers. We utilize household-level sample
weights to produce nationally-representative estimates.3
Our baseline estimates use monthly cash income values (which includes cash assistance)
and reported family size. Table 1 also reports alternative estimates using quarterly income and an
equivalence scale.4 Ideally we would also report annual estimates. Unfortunately, to produce
annual estimates, the SIPP requires the use of calendar year weights, which have not yet been
made available for the final year of our study period (They are also not available for some years
in the study period because of breaks between SIPP panels). Also, monthly estimates better
protect against biasing from non-random attrition throughout SIPP panels. Still, this represents
an important limitation of the current analysis. It is possible that households could experience a
month or even a calendar quarter in extreme poverty, but have larger incomes over a full year.
For monthly estimates, we keep only SIPP reporting month (4th reference month)
observations in each wave, because these are known to be the most accurate (Moore, 2007).5 We
replicate results with all SIPP months (see table 3), and findings are substantively similar. As is
commonly done, reporting-month observations within each wave are stacked to generate crosssectional estimates. For our quarterly estimates, we use reporting month observations and
average months across two waves (reference month 3 wave t + reference month 4 wave t +
3
Our full sample consists of 12,244 unweighted household observations reporting extreme poverty using the
baseline definition, starting with 256 household observations in wave 1 of the 1996 panel, and ending with 392
household observations in wave 9 of the 2008 panel.
4
The square root of household size.
5
Sensitivity analyses reported in table 3 show that results are consistent when using all SIPP months.
9
reference month 1 wave t+1). To minimize bias caused by non-random attrition, we do not want
to drop cases not present in wave t+1. In those cases we use reference months 2, 3, and 4 of wave
t.
There are concerns that survey respondents may under-report income, both from work
and public program participation. However, when compared to other large nationally
representative peer surveys, studies find that the SIPP is generally superior at measuring income
among the poor and public program participation (Czajka & Denmead, 2008; Meyer, Mok, &
Sullivan, 2008). Czajka & Denmead (2008) estimate aggregate household income in the U.S.
across a number of major household surveys. The SIPP generates an estimate of $391 billion in
income among families in the lowest income quintile, while the Current Population Survey
(CPS) estimate is $371 billion and the American Community Survey estimate is $369 billion.
Further, administrative earnings data (such as from unemployment insurance records or IRS tax
records) are insufficient for capturing informal income among the poor. Thus, the SIPP is the
source of nationally representative data that records the largest amount of income among the
poor, making it the most appropriate choice for the current study.
In terms of capturing program participation, the SIPP also performs better, on average,
than CPS and other major household surveys, although it is far from perfect. Meyer et al.
compare reports of public program participation in a number of major household surveys to
administrative caseload totals. The SIPP’s reporting rate for SNAP was 84.2 percent in 1996 and
remained relatively stable through 2005, the final year Meyer et al. report on. In contrast, the
CPS’s reporting rate for SNAP was 66.3 in 1996 and fell to 56.5 percent in 2005. For
AFDC/TANF, the SIPP reporting rate started at 79.5 percent in 1996, fell to 65.5 percent in 2002,
but increased to 80.9 percent by 2005. For CPS, the AFDC/TANF reporting rate started at 67.0
10
percent in 1996, fell below 60 percent in a number of years, and ended at 63.0 percent in 2005
(and even down to 52.7 in 2007). It is worth noting that the SIPP reporting rates do not
consistently worsen across the study period, at least up to 2005. Thus, falling reporting rates over
time cannot explain an increase in the prevalence of extreme poverty over the study period.
We report three estimates of extreme poverty, using three different definitions of
household income. The first considers only reported household cash income from all sources, as
described above. This definition most closely aligns with the official U.S. poverty measure. The
second adds SNAP benefits, (assuming $1 SNAP = $1 cash). The third accounts for SNAP and
also adds the estimated average monthly value of the household’s net refundable tax credits,
specifically, the EITC and Child Tax Credit, the value of which is often greater than the income
tax liability incurred by recipients, and government housing subsidies (section 8; public housing).
Definitions 2 and 3 more closely align with the new supplemental poverty measure by taking
into account major public programs.
Any household reporting a housing subsidy is considered to be above the cut-off for
extreme poverty.6 To estimate the total value of refundable federal tax credits, we estimate the
average monthly household earned and unearned income reported by the household head for the
full calendar year of the observation, and then scale these monthly estimates up to equivalent
annual earned and unearned household income estimates. This allows us to include observations
that are in the sample for less than 12 months, which is important because of non-random
attrition and panel breaks. We use TAXSIM to generate federal income and FICA (payroll) tax
liabilities. Among low-income working households with children, annual federal income tax
6
Halfway through 2011, roughly one-in-five households in extreme poverty utilized a housing subsidy such as
section 8 vouchers or public housing units (comparable to the proportion of all households in poverty). Additionally,
about two thirds of these households had at least one child covered by public health insurance, somewhat less than
the same proportion for all households in poverty.
11
liabilities will be a net negative (indicating a transfer from the government to the household)
because the value of refundable credits will exceed income tax liability. However, FICA taxes
are regressive and for the same family will result in a transfer from the household to the
government). Thus, we take each household’s refunded income tax liability, less FICA,
(censoring at zero because we do not want to code someone as being in extreme poverty as a
result of annual FICA liabilities), and divide this value by 12 to create a monthly amount. This
value is added to each respondent’s total monthly household income value by year.
Results
Figure 1 plots the number of households with non-elderly heads and minor children
living in extreme poverty between 1996 and 2011. Breaks in the lines represent breaks between
SIPP panels. We estimate that the number of households living on $2 or less in cash income per
person, per day in a given month increased from about 636,000 in 1996 to about 1.65 million in
mid-2011, a percentage growth of 159.1 percent. According to these estimates, in mid-2011
about 3.55 million children lived in extreme poverty each month (see Table 2).
Figure 1 shows that counting means-tested benefits as income reduces—but does not
eliminate—this rise in extreme poverty. With SNAP included as income, the number of extremepoor households increases by 80.4 percent, from roughly 475,000 to 857,000. When federal
refundable tax credits and housing subsidies are accounted for, in addition to SNAP, the increase
is about 50 percent, from 409,000 to 613,000 households, which included 1.17 million children
(see table 2). Virtually all of the difference between the SNAP-only trend line and this final one
is attributable to refundable tax credits.
The beneficial effects of these means-tested public programs appear especially notable
after the passage of the ARRA. As figure 1 demonstrates, extreme poverty based only on cash
12
income increased by about 35 percent between late 2007 and early 2011. But when SNAP
benefits and refundable tax credits and housing subsidies are added, the rise in extreme poverty
was less than 10 percent over this period, in fact falling slightly in some months.
A final trend line presented in figure 1 subtracts income from cash assistance provided
through AFDC/TANF from the cash income only estimate. So this line presents the number of
households with children living on less than $2 in cash income, per person, per day in a given
month, not accounting for cash assistance. In 1996, cash assistance had a substantial effect on
extreme poverty, bringing the incomes of 1.15 million households above the extreme poverty
threshold. Throughout the 1990s, the impact of cash assistance in reducing extreme poverty
declined substantially, and flattened out in the early 2000s. By mid-2011, cash assistance was
lifting the incomes of only about 291,000 families above the extreme poverty threshold.
Figure 2 examines the trend in extreme poverty as a proportion of all households with
children. Extreme poverty grew steadily as a proportion of all households over the study period,
spiking during the Great Recession. In 1996, households in extreme poverty represented about
1.7 percent of all households. Between 1997 and 2000—a period of low unemployment—this
proportion grew to about 2.0 percent. Between 2001 and 2010, this incidence grew from 2.3
percent to about 3.0 percent, and hovered between 4.0 and 4.3 percent through mid-2011.
When SNAP is included, households in extreme poverty rise from about 1.3 percent of all
households with children in 1996 to 2.2 percent in 2011, an increase of 70.5 percent. Adding
refundable tax credits and housing subsidies reduces the percentage increase to about 44.1
percent. Finally, the trend line omitting cash assistance from AFDC/TANF follows a similar
pattern as the corresponding trend line in figure 1. In 1996, cash assistance was lifting between
13
2.5 and 3.1 percent of all households with children out of extreme poverty every month. By mid2011, it was lifting only about 0.75 percent of families out of extreme poverty.
Table 1 presents some alternative estimates of extreme poverty. First, instead of dividing
income by the number of persons in the household, we used an equivalence scale (the square root
of household size). This reduces the extreme poverty count in mid-2011, from 1.65 to 1.40
million, but the rate of growth is consistent with our baseline estimates.
We also averaged each household’s income over three months, a quarterly estimate.
Fewer households experience extreme poverty for a calendar quarter, when compared to a month.
However, the percentage growth in extreme poverty based on quarterly income is consistent with
the monthly estimates. After adjusting for means-tested programs, the growth in the number of
households experiencing extreme poverty was 49.9 percent using our monthly measure. In
contrast, it was 97.4 percent using quarterly income, and even higher when an equivalence scale
is used. These higher growth rates are likely a function of lower starting values in 1996. Still, it
appears that the proportion of households with children exposed to extreme poverty has grown
most dramatically, even when longer rather than shorter spells are considered.
The bottom row of Table 1 reports a final specification for our baseline definition (using
cash income only). The number of households with children who reported at least 1 month of
extreme poverty over a calendar quarter increased from about 1.30 million in 1996 to 2.41
million households in mid-2011, consisting of 6.3 percent of all households with children in mid2011.
Table 2 looks at the characteristics of the households living in extreme poverty using first
cash income only, and then adjusting for means-tested program participation. These estimates,
particularly, should be treated with caution because different sub-groups may have different rates
14
of under-reporting of program participation and income, which are averaged out in the
population-level estimates. Such variable rates of under-reporting could bias estimates of relative
trends across groups. Still, these estimates may prove insightful for understanding which subgroups have been most impacted by the rise in extreme poverty, and whether these findings are
consistent with the possibility that at least part of this rise is related to changes stemming from
the 1996 welfare reform. Using the cash only definition, in 2011 about 36.5 percent of the
households in extreme poverty were headed by a married couple, and 50.8 percent were headed
by a single female. After adjusting for means-tested programs, just over half of the extreme
poverty households are married, and less than one-third are single female headed, reflecting
greater reliance on public programs among single female headed households.
In mid-2011, using the cash only definition, about 47.5 percent of households in extreme
poverty were headed by white non-Hispanics, and 46.0 percent were headed by African
Americans or Hispanics (reported together because of small sample sizes). After adjusting for
means-tested program participation, the proportion headed by white non-Hispanics increases to
about 61.2 percent. Thus, extreme poverty is not limited to households headed by single mothers
or disadvantaged minorities. Still, it is clear that the percentage growth in extreme poverty over
our study period was greatest among vulnerable groups who were most likely to be impacted by
the 1996 welfare reform. The percentage growth in extreme poverty for households headed by
single females was 229.9 percent using cash only, and 67.9 percent after adjusting for meanstested programs. Racial minorities saw a larger growth in extreme poverty under the cash only
definition, although they may have been better buffered against the rise in extreme poverty by
major means-tested income programs, when compared to non-Hispanic whites.
15
One concern is that these results are being driven by rising rates of imputed values for
income and program participation variables in the SIPP, over time (Meyer et al., 2008).
Imputation rates are consistently low in the first wave of each panel, rise throughout waves
within panels, and are typically at their highest level in the final waves of a panel. Thus, if rising
rates of imputation are driving the substantive findings presented here, this should generate
significant declines in extreme poverty between the final wave of one panel (with high rates of
imputation), and the first wave of the next (with low imputation). This can be directly examined
by looking at the estimates across the breaks between panels in figures 1 and 2. In most cases,
the estimate for extreme poverty increases in the first wave of a new panel (when imputation is
low), relative to the last wave of the previous one (when imputation is high). Thus, it appears
unlikely that rising rates of imputation are driving the rise in extreme poverty over the study
period.
Another concern is that the rise in extreme poverty is being driven by other factors
besides the 1996 welfare reform. It is beyond the scope of the current study to fully assess the
casual mechanisms of rising extreme poverty in the U.S. Still, some analyses may strengthen the
case that the retrenchments enacted through the 1996 reform are at least partially to blame for
this trend. One compelling piece of evidence that this is the case is provided in the final trend
lines in figures 1 and 2 (cash income – AFDC/TANF), which have been previously discussed. It
is obvious that cash assistance was having a substantial impact in reducing extreme poverty in
early 1996, and having a much smaller effect in reducing extreme poverty by mid-2011.
Another test examines the trends of a comparison group. If the increase in extreme
poverty is equally as large for households without children—who, as a group, have never been
eligible for cash assistance—as it is for households with children, this would call into question
16
the extent to which the 1996 welfare reform has driven the rise in extreme poverty. Table 3
presents the starting points and ending points for the proportion of households without children
in extreme poverty during our study period. While this group does experience an increase in
extreme poverty over the study period, the growth is less than half of the percentage growth in
extreme poverty experienced by households with children.7 This is consistent with a hypothesis
that some part of the rise in extreme poverty can be attributed to the welfare reform, and some
portion of the rise is related to other causes, such as rising unemployment during the Great
Recession. As expected, table 3 further shows that means-tested programs are having a much
greater impact in reducing extreme poverty among households with children. In mid-2011,
means-tested programs reduced the proportion of households with children in extreme poverty
from 4.3 percent to 1.6 percent. But for households without children, means-tested programs
reduced the proportion in extreme poverty from 5.1 percent to 4.3 percent (and virtually all of
this reduction was due to SNAP).
Finally, it is important to consider how the estimates presented here are impacted by the
exact threshold used for the extreme poverty measure. If the substantive finding of rising
extreme poverty is being driven by the retrenchment of cash assistance at the very bottom of the
income distribution, than we would expect this trend of rising extreme poverty—based on cash
income—to dissipate gradually as the threshold is increased from $2 a day to higher levels, and
labor market participation becomes more common. That is to say, as the threshold, by definition,
includes more individuals with other sources of income, the impact of the retrenchment of cash
assistance will be more muted. Table 3 reports on estimates using three alternative thresholds:
<=$4.00 per person, per day; <=$6.00; <=$8.00. When looking at the cash only definition, the
7
In addition, examining the full trend line for households without children shows that there was no increase in
extreme poverty for this group during the 1990s, whereas households with children saw a substantial increase.
17
estimates presented in table 3 follow the pattern as expected. As the threshold is increased, the
rise over time in the proportion of families below that threshold gradually declines, as compared
to the baseline estimate of $2.00 per person, per day.8
Conclusion
As of mid-2011, about 1.65 million households with about 3.55 million children were
surviving on $2 or less in cash income per person per day in a given month. These estimates
account for income received from TANF and other direct cash income transfer programs.
Households in extreme poverty constituted 4.3 percent of all non-elderly households with
children. The prevalence of extreme poverty rose sharply between 1996 and 2011, with the
highest growth rates found among groups most affected by the 1996 welfare reform. When
income over the quarter is used rather than income from a single month, the proportional
increase in extreme poverty over the study period is comparable the monthly estimates (and in
some cases larger), although the overall incidence is lower.
When we consider SNAP benefits as equivalent to dollars, this reduces the number of
extremely poor households with children in mid-2011 by 48.0 percent, and when refundable tax
credits and housing subsidies are subsequently added, the number falls by 62.8 percent. We
estimate that these major means-tested aid programs currently save roughly 2.38 million children
from extreme poverty each month, but leave 1.17 million children behind. These estimates come
from a household survey and thus fluctuate somewhat from month-to-month, as is made obvious
in figures 1 and 2. Therefore, the exact point estimates should be treated with caution. Of more
interest is the trend over time, which is quite clear, substantial, and robust to a number of
sensitivity tests.
8
In fact, the growth in the proportion of households with children living at or below $8.00 per person per day is
quite comparable to official estimates of the growth in the number and proportion of families with children in deep
poverty between 1995 and 2011.
18
Our estimates may be biased by under-reporting of income by SIPP respondents. While
this is a limitation of our work, the SIPP does comparatively well relative to other major surveys
in terms of reporting rates (Meyer, Mok & Sullivan, 2009), and reporting rates for most
programs have not fallen steadily over our study period in a way that would explain the dramatic
and steady increase in extreme poverty we report. Further, under-reporting of income itself
suggests adverse outcomes, such as engagement in the underground economy (Edin & Lein,
1997). For example, 8 percent of Edin & Lein’s sample of welfare-reliant mothers in their study
Making Ends Meet reported engaging in underground work, most commonly selling sex. If
under-reporting of income explains the rising prevalence of extreme poverty in the U.S., then
this may be a cause of serious concern. Another possibility is that, in some cases workers at a job
who are “off the books” may report income in the SIPP, but may not be able to use it to claim
refundable tax credits. This may lead to an upward bias of the estimated impact of the EITC and
other refundable tax credits on extreme poverty.
The descriptive analyses presented here cannot clarify the exact causal mechanisms that
have led to such a sharp uptick in extreme poverty in the U.S. We hypothesize that the virtual
disappearance of a cash safety net for non-workers has played an important role, combined with
overall slow economic growth during the 2000s, culminating in the major job losses of the Great
Recession. However, the primary purpose of the current descriptive analyses is to document the
rise in extreme poverty over the past 15 years, show the impact that means-tested income transfer
programs are having on this trend, and to call for more research. According to our estimates, a
growing population of children experience spells with virtually no income, and more research,
both quantitative and qualitative, is needed to examine this. Future research should more fully
account for the causal mechanisms that have led to this uptick in extreme poverty, and further
19
build knowledge regarding what circumstances at the household level tend to precipitate a spell
of extreme poverty, and how households survive during such spells.
It is clear that our current major safety-net programs are playing a vital role at the very
bottom, especially in the aftermath of the Great Recession, and are blunting some of the hardship
that these households would otherwise face. However, it would be wrong to conclude that the
U.S. safety net is strong, or even adequate, when the number and proportion of households with
children surviving on less than $2 per day has risen so dramatically over the past 15 years, even
after accounting for means-tested transfers.
20
References
Appelbaum, B. & Gebeloff, R. (February 11, 2012). “Even Critics of the Safety Net Increasingly
Depend on It.” The New York Times.
Ben-Shalom, Y., Moffit, R., & Scholz, J.K. (in press). An assessment of the effectiveness of antipoverty programs in the United States. In Oxford Handbook of Economics of Poverty, P.
Jefferson, ed. New York: Oxford University Press. Currently available as NBER working
paper No. w17042, available at: http://www.nber.org/papers/w17042.
Blank, R.M. (2007). Improving the safety net for single mothers who face serious barriers to
work. The Future of Children, 17(2), 183-197.
Czajka, J.L., & Denmead, G. (2008). “Income Data for Policy Analysis: A Comparative
Assessment of Eight Surveys. Final Report.” Report by Mathematica Policy Research,
Inc. under contract to the Department of Health and Human Services, ASPE. Retrieved
from the web 11/1/2009, http://www.mathematicampr.com/publications/PDFs/incomedata.pdf.
Edin, K. & Lein, L. (1997). Making ends meet. New York: Russell Sage Foundation.
Gould-Werth, A. & Shaefer, H.L. (2012). Unemployment insurance participation by education
and by race and ethnicity. Monthly labor review, October, 28-41.
Loprest, P.J. (November 2011). “Disconnected families and TANF”. Urban Institute, OPRE
research brief ,#02.
Mendenhall, R., Edin, K., Crowley, S., Sykes, J., Tach, L., Kriz, K. & Kling, J. (2012). The role
of Earned Income Tax Credit in the budgets of low-income families. Social service
review, 86(3), 367-400.
Meyer, B.D., Mok, W.K.C., & Sullivan, J.X. (2009). The under-reporting of transfers in
household surveys: Its nature and consequences, National Bureau of Economic Research
Working Paper No. 15181.
O’Leary, C. J. & Kline, K. J. (2008), UI as a safety net for former TANF recipients, ASPE
Project: HS-05-001, submitted to the US Department of Health and Human Services
ASPE.
Ravallion, M., Chen, S. & Sangraula, P. (2008). Dollar a day revisited. The World Bank
Development Research Group Policy Research Working Paper 4620. Retrieved from the
the web 1/22/2013, http://wwwwds.worldbank.org/external/default/WDSContentServer/IW3P/IB/2008/09/02/00015834
9_20080902095754/Rendered/PDF/wps4620.pdf.
Shaefer, H.L. (20120). Identifying key barriers to Unemployment Insurance for disadvantaged
workers in the United States. Journal of social policy, 39(3), 439-460.
Sherman, A., Greenstein, R. & Ruffing, K. (2012). “Contrary to ‘Entitlement Society’ Rhetoric,
Over Nine-Tenths of Entitlement Benefits Go to Elderly, Disabled, or Working
Households. Report by the Center on Budget and Policy Priorities. Retrieved from the
web 2/16/2012, http://www.cbpp.org/files/2-10-12pov.pdf.
21
Tax Policy Center, (2010). “Tax Facts on Historical EITC,” Tax Policy Center, Washington, DC,
http://www.taxpolicycenter.org/taxfacts/displayafact.cfm?Docid=37.
Turner, L.J., Danziger, S. & Seefeldt, K.S. (2006). Failing the transition from welfare to work:
Women chronically disconnected from employment and cash welfare. Social science
quarterly, 87(2), 227-249.
Weiner, R. (February 1, 2012). “Romney, citing safety net, says he’s ‘not concerned about the
very poor’.” The Washington Post.
22
Figure 1: Households with Children in Extreme
Poverty in the United States
(<=$2 per person, per day)
1,800,000
1,600,000
1,400,000
1,200,000
1,000,000
800,000
600,000
400,000
200811
200907
201003
201011
201107
200408
200504
200512
200608
200704
200108
200204
200212
200308
0
199606
199702
199710
199806
199902
199910
200,000
Cash Income (Includes AFDC/TANF)
Cash Income + SNAP
Cash Income + SNAP, Tax Credits, Housing Subsidies
Cash Income - AFDC/TANF
Source: Authors' analyses of the 1996 through 2008 Panels of the SIPP. The horizontal axis represents
approximate years and months of SIPP 4th reference month estimates. The vertical axis represents the
number of non-elderly households with children. Breaks in the trend lines represent breaks in the SIPP
panels.
23
Figure 2: Proportion of Households with Children
in Extreme Poverty
(<=$2 per person, per day)
5.00%
4.00%
3.00%
2.00%
200811
200907
201003
201011
201107
200408
200504
200512
200608
200704
200108
200204
200212
200308
0.00%
199606
199702
199710
199806
199902
199910
1.00%
Cash Income (Includes AFDC/TANF)
Cash Income + SNAP
Cash Income + SNAP, Tax Credits, Housing Subsidies
Cash Income - AFDC/TANF
Source: Authors' analyses of the 1996 through 2008 Panels of the SIPP. The horizontal axis represents
approximate years and months of SIPP 4th reference month estimates. The vertical axis represents the
proportion of non-elderly households with children. Breaks in the trend lines represent breaks in the SIPP
panels.
24
TABLE 1: ALTERNATIVE ESTIMATES OF EXTREME POVERTY OF HOUSEHOLDS WITH CHILDREN
Baseline monthly estimates (Figs. 1 and 2)
Cash only
Add SNAP
Add SNAP, tax credits, housing subsidies
Monthly with equivalence scale
Cash only
Add SNAP
Add SNAP, tax credits, housing subsidies
Number of households with children
(Rounded in thousands)
1996
2011
% Growth
Percent of households
with children
1996 2011 % Growth
636,000
475,000
409,000
1,648,000
857,000
613,000
159.1%
80.4%
49.9%
1.7% 4.3%
1.3% 2.2%
1.1% 1.6%
151.8%
69.2%
45.5%
507,000
408,000
293,000
1,401,000
732,000
476,000
176.3%
79.4%
62.5%
1.4% 3.6%
1.1% 1.9%
0.8% 1.2%
157.1%
72.7%
50.0%
307,000
209,000
189,000
1,039,000
478,000
373,000
238.4%
128.7%
97.4%
0.8% 2.7%
0.6% 1.2%
0.5% 1.0%
237.5%
100.0%
100.0%
212,000
167,000
136,000
797,000
360,000
273,000
275.9%
115.6%
100.7%
0.6% 2.1%
0.5% 0.9%
0.4% 0.7%
250.0%
80.0%
75.0%
85.9%
3.5% 6.3%
80.0%
Quarterly estimates
Cash only
Add SNAP
Add SNAP, tax credits, housing subsidies
Quarterly with equivalence scale
Cash only
Add SNAP
Add SNAP, tax credits, housing subsidies
1 or more months per quarter (cash only)
1,295,000 2,407,000
Source: Authors' Analyses of the 1996 and 2008 Panels of the SIPP
25
TABLE 2: CHARACTERISTICS OF HOUSEHOLDS WITH CHILDREN IN EXTREME POVERTY
Number in extreme poverty, monthly
Cash Only
Adjusted for Means-tested Programs
1996
2011
% Growth
1996
2011
% Growth
Total households
636,000
1,648,000
159.1%
409,000
613,000
49.9%
Married households
323,000
602,000
86.4%
255,000
330,000
29.4%
Single female households
254,000
838,000
229.9%
112,000
188,000
67.9%
Race of household head
White, Non-Hispanic
African American & Hispanic
334,000
265,000
782,000
758,000
Children
1,383,000
3,547,000
Source: Authors' Analyses of the 1996 and 2008 Panels of the SIPP
134.1%
186.0%
243,000
130,000
375,000
182,000
54.3%
40.0%
156.5%
788,000
1,166,000
48.0%
TABLE 3: HOUSEHOLDS IN EXTREME POVERTY: SENSITIVITY ANALYSES
Percent of Households
Households with children (Baseline)
Using all months of data
1996
1.7%
1.9%
Cash Only
2011
% Growth
4.3%
152.9%
4.0%
110.5%
Adjusted for Means-tested
Programs
1996
2011
% Growth
1.1%
1.6%
45.5%
1.0%
1.7%
70.0%
Households without children
3.0%
5.1%
70.0%
2.6%
4.3%
65.4%
5.7%
7.5%
9.6%
83.9%
44.2%
31.5%
1.5%
2.2%
3.3%
2.6%
3.8%
4.9%
73.3%
72.7%
48.5%
Households with children
Using <=$4 threshold 3.1%
Using <=6 threshold 5.2%
Using <=$8 threshold 7.3%
Source: Authors' Analyses of the 1996 and 2008 Panels of the SIPP
26